A key requirement for RRAM in neural network accelerators with a large number of synaptic parameters is the multilevel programming. This is hindered by resistance imprecision due to cycle-to-cycle and device-to-device variations. Here, we compare two multilevel programming algorithms to minimize resistance variations in a 4-kbit array of HfO2 RRAM. We show that gate-based algorithms have the highest reliability. The optimized scheme is used to implement a neural network with 9-level weights, achieving 91.5% (vs. software 93.27%) in MNIST recognition.
Optimized programming algorithms for multilevel RRAM in hardware neural networks
Zambelli C.;Olivo P.;
2021
Abstract
A key requirement for RRAM in neural network accelerators with a large number of synaptic parameters is the multilevel programming. This is hindered by resistance imprecision due to cycle-to-cycle and device-to-device variations. Here, we compare two multilevel programming algorithms to minimize resistance variations in a 4-kbit array of HfO2 RRAM. We show that gate-based algorithms have the highest reliability. The optimized scheme is used to implement a neural network with 9-level weights, achieving 91.5% (vs. software 93.27%) in MNIST recognition.File in questo prodotto:
| File | Dimensione | Formato | |
|---|---|---|---|
|
irps_2021_manuscript_final.pdf
solo gestori archivio
Descrizione: Pre-print
Tipologia:
Pre-print
Licenza:
NON PUBBLICO - Accesso privato/ristretto
Dimensione
2.77 MB
Formato
Adobe PDF
|
2.77 MB | Adobe PDF | Visualizza/Apri Richiedi una copia |
|
Optimized_programming_algorithms_for_multilevel_RRAM_in_hardware_neural_networks.pdf
solo gestori archivio
Descrizione: Full text editoriale
Tipologia:
Full text (versione editoriale)
Licenza:
NON PUBBLICO - Accesso privato/ristretto
Dimensione
2.85 MB
Formato
Adobe PDF
|
2.85 MB | Adobe PDF | Visualizza/Apri Richiedi una copia |
I documenti in SFERA sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
